Novel deep learning approach to estimate rigid gas permeable contact lens base curve for keratoconus fitting.

INTRODUCTION: Rigid gas permeable contact lenses (RGP) are the most efficient means of providing optimal vision in keratoconus. RGP fitting can be challenging and time-consuming for ophthalmologists and patients. Deep learning predictive models could simplify this process. OBJECTIVE: To develop a deep learning model to predict the base curve (R0) of rigid gas permeable contact lenses for keratoconus patients. METHODS: We conducted a retrospective study at the Rothschild Foundation Hospital between June 2012 and April 2021. We included all keratoconus patients fitted with Menicon Rose K2® lenses. The data was divided into a training set to develop the model and a test set to evaluate the model's performance. We used a U-net architecture. The raw matrix of anterior axial curvature in millimeters was extracted from Scheimpflug examinations for each patient and used as input for the model. The mean absolute error (MAE) between the prediction and the prescribed R0 was calculated. Univariate and multivariate analyses were conducted to assess the model's errors. RESULTS: Three hundred fifty-eight eyes from 202 patients were included: 287 eyes were included in the training dataset, and 71 were included in the testing dataset. Our model's Pearson coefficient of determination (R2) was calculated at 0.83, compared to 0.75 for the manufacturer's recommendation (mean keratometry, Km). The mean square error of our model was calculated at 0.04, compared to 0.11 for Km. The predicted R0 MAE (0.16 ± 0.13) was statistically significantly different from the Km MAE (0.23 ± 0.23) (p = 0.02). In multivariate analysis, an apex center outside the central 5 mm region was the only factor significantly increasing the prediction absolute error. CONCLUSION: Our deep learning approach demonstrated superior precision in predicting rigid gas permeable contact lens base curves for keratoconus patients compared to the manufacturer's recommendation. This approach has the potential to be particularly beneficial in complex fitting cases and can help reduce the time spent by ophthalmologists and patients during the process.

Discrimination between keratoconus, forme fruste keratoconus, and normal eyes using a novel OCT-based tomographer.

PURPOSE: To combine objective machine-derived corneal parameters obtained with new swept-source optical coherence tomography (SS-OCT) tomographer (Anterion) to differentiate between normal (N), keratoconus (KC) and forme fruste KC (FFKC). SETTING: Laser Center, Hôpital Fondation Adolphe de Rothschild, Paris, France. DESIGN: Retrospective study. METHODS: 281 eyes of 281 patients were included and divided into 3 groups: N (n = 156), FFKC (n = 43), and KC (n = 82). Eyes were included in each group based on objective evaluation using Nidek Corneal Navigator, and subjective evaluation by authors. The SS-OCT system provided anterior and posterior corneal surface and pachymetry derived variables. The training set was composed of 143 eyes (95 N, 43 FFKC). Discriminant analysis was used to determine the group of an observation based on a set of variables. The obtained formula was tested in the validation set composed of 61 N and 82 KC. RESULTS: Among curvature parameters, the FFKC had significantly higher irregularity index at 3 mm and 5 mm, higher inferior-superior index, higher SteepK-OppositeK index and inferiorly decentered posterior steepest keratometry. Among thickness parameters: central pachymetry, thinnest pachymetry, percentage of thickness increase from center to periphery, and inferior decentration of the thinnest point were statistically different between groups. Combination of multiple variables into a discriminant function (F1) included 5 parameters and reached an area under the receiver operating characteristic curve (AUROC) of 0.95 (sensitivity = 75%, specificity = 98.5%) for detection of FFKC. F1 differentiates N from KC with AUROC = 0.99 (sensitivity = 99%, specificity = 99%). CONCLUSIONS: Combining anterior and posterior curvatures variables along with pachymetric data obtained from SS-OCT allowed automated detection of early KC and KC with very good accuracy (87% and 99.5% respectively).

Descemet Membrane Endothelial Keratoplasty-Induced Refractive Shift and Descemet Membrane Endothelial Keratoplasty-Induced Intraocular Lens Calculation Error.

PURPOSE: The aim of this study was to determine the mechanisms leading to the refractive shift and intraocular lens calculation error induced by Descemet membrane endothelial keratoplasty (DMEK), using ocular biometry and corneal elevation tomography data. METHODS: This is a retrospective, monocentric cohort study. Eyes which underwent uncomplicated DMEK surgery with available pre-DMEK and post-DMEK Scheimpflug rotating camera data (Pentacam, Oculus, Wetzlar, Germany) were considered for inclusion with an age-matched control group of healthy corneas. Cataract surgery data were collected for triple-DMEK cases. DMEK-induced refractive shift (DIRS) and intraocular lens calculation error (DICE) were calculated. Pearson r correlation coefficient was calculated between each corneal parameter variation and both DIRS and DICE. RESULTS: DIRS was calculable for 49 eyes from 43 patients. It was 30.61% neutral, 53.06% hyperopic (36.73% > 1D), and 16.32% myopic (6.12% > 1 D). DICE was calculable for 30 eyes of 26 patients: It was 46.67% neutral, 40.00% hyperopic (10.00% > 1D), and 13.33% myopic (3.33% > 1D). DIRS and DICE were mainly associated with variations in PRC/ARC ratio, anterior average radii of curvature (ARC), posterior average radii of curvature (PRC), and posterior Q. CONCLUSIONS: Our results suggest that ARC variations, PRC/ARC ratio variations, PRC variations, and posterior Q variations are the most influential parameters for both DIRS and DICE. We suggest that a distinction between those different phenomenons, both currently described as "hyperopic shift" in the literature, should be made by researchers and clinicians.

Deep Learning Model for the Detection of Corneal Edema Before Descemet Membrane Endothelial Keratoplasty on Optical Coherence Tomography Images.

Purpose: Descemet membrane endothelial keratoplasty (DMEK) is the preferred method for treating corneal endothelial dysfunction, such as Fuchs endothelial corneal dystrophy (FECD). The surgical indication is based on the patients' symptoms and the presence of corneal edema. We developed an automated tool based on deep learning to detect edema in corneal optical coherence tomography images. This study aimed to evaluate this approach in edema detection before Descemet membrane endothelial keratoplasty surgery, for patients with or without FECD. Methods: We used our previously described model allowing to classify each pixel in the corneal optical coherence tomography images as "normal" or "edema." We included 1992 images of normal and preoperative edematous corneas. We calculated the edema fraction (EF), defined as the ratio between the number of pixels labeled as "edema," and those representing the cornea for each patient. Differential central corneal thickness (DCCT), defined as the difference in central corneal thickness before and 6 months after surgery, was used to quantify preoperative edema. AUC of EF for the edema detection was calculated for Several DCCT thresholds and a value of 20 µm was selected to define significant edema as it provided the highest area under the curve value. Results: The area under the curve of the receiver operating characteristic curve for EF for the detection of 20 µm of DCCT was 0.97 for all patients, 0.96 for Fuchs and normal only and 0.99 for non-FECD and normal patients. The optimal EF threshold was 0.143 for all patients and patients with FECD. Conclusions: Our model is capable of objectively detecting minimal corneal edema before Descemet membrane endothelial keratoplasty surgery. Translational Relevance: Deep learning can help to interpret optical coherence tomography scans and aid the surgeon in decision-making.

Development and validation of a pixel wise deep learning model to detect cataract on swept-source optical coherence tomography images.

PURPOSE: The diagnosis of cataract is mostly clinical and there is a lack of objective and specific tool to detect and grade it automatically. The goal of this study was to develop and validate a deep learning model to detect and localize cataract on Swept Source Optical Coherance Tomography (SS-OCT) images. METHODS: We trained a convolutional network to detect cataract at the pixel level from 504 SS-OCT images of clear lens and cataract patients. The model was then validated on 1326 different images of 114 patients. The output of the model is a map repreenting the probability of cataract for each pixel of the image. We calculated the Cataract Fraction (CF), defined as the number of pixel classified as "cataract" divided by the number of pixel representing the lens for each image. Receiver Operating Characteristic Curves were plotted. Area Under the Curve (ROC AUC) sensitivity and specitivity to detect cataract were calculated. RESULTS: In the validsation set, mean CF was 0.024 ± 0.077 and 0.479 ± 0.230 (p < 0.001). ROC AUC was 0.98 with an optimal CF threshold of 0.14. Using that threshold, sensitivity and specificity to detect cataract were 94.4% and 94.7%, respectively. CONCLUSION: We developed an automatic detection tool for cataract on SS-OCT images. Probability maps of cataract on the images provide an additional tool to help the physician in its diagnosis and surgical planning.

Corneal Edema Visualization With Optical Coherence Tomography Using Deep Learning: Proof of Concept.

PURPOSE: Optical coherence tomography (OCT) is essential for the diagnosis and follow-up of corneal edema, but assessment can be challenging in minimal or localized edema. The objective was to develop and validate a novel automated tool to detect and visualize corneal edema with OCT. METHODS: We trained a convolutional neural network to classify each pixel in the corneal OCT images as "normal" or "edema" and to generate colored heat maps of the result. The development set included 199 OCT images of normal and edematous corneas. We validated the model's performance on 607 images of normal and edematous corneas of various conditions. The main outcome measure was the edema fraction (EF), defined as the ratio between the number of pixels labeled as edema and those representing the cornea for each scan. Overall accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve were determined to evaluate the model's performance. RESULTS: Mean EF was 0.0087 ± 0.01 in the normal scans and 0.805 ± 0.26 in the edema scans (P < 0.0001). Area under the receiver operating characteristic curve for EF in the diagnosis of corneal edema in individual scans was 0.994. The optimal threshold for distinguishing normal from edematous corneas was 6.8%, with an accuracy of 98.7%, sensitivity of 96.4%, and specificity of 100%. CONCLUSIONS: The model accurately detected corneal edema and distinguished between normal and edematous cornea OCT scans while providing colored heat maps of edema presence.

Assessment of corneal epithelial thickness mapping in epithelial basement membrane dystrophy.

PURPOSE: To investigate the corneal epithelial thickness topography with optical coherence tomography (OCT) and its relationship with vision quality in epithelial basement membrane dystrophy (EBMD). METHODS: 45 eyes of EBMD patients, 26 eyes of dry eye (DED) patients and 22 eyes of normal subjects were enrolled. All participants were subjected to 9-mm corneal epithelial mapping with OCT and vision quality was assessed with the optical quality analysis system using the objective scatter index (OSI). Central, superior, inferior, minimum, maximum, and standard deviation of epithelium thickness (Irregularity), were analysed and correlations with the OSI were calculated. RESULTS: The mean (±SD) central, inferior and maximum epithelial thicknesses of the EBMD patients (respectively, 56.4 (±8.1) µm, 58.9 (±6.4) µm, and 67.1 (±8.3) µm) were thicker compared to DED patients (P<0.05) and normal subjects (P<0.05). We found greater irregularity of epithelial thickness in EBMD (5.1±2.5 µm) compared to DED patients (2.6±1.0 µm) (P = 4.4.10-6) and normal subjects (2.1±0.7 µm) (P = 7.6.10-7). The mean OSI was worse in EBMD patients than in DED patients (P = 0.01) and compared to normal subjects (P = 0.02). The OSI correlated with the epithelial thickness irregularity (Spearman coefficient = 0.54; P = 2.65.10-5). CONCLUSIONS: The OCT pachymetry map demonstrated that EBMD patients had thicker corneal epithelium in the central and inferior region. These changes were correlated with objective measurements of vision quality. This OCT characterisation of the EMBD provides a better understanding of the epithelial behaviour in this dystrophy and its role in vision quality.

Unsupervised learning for large-scale corneal topography clustering.

Machine learning algorithms have recently shown their precision and potential in many different use cases and fields of medicine. Most of the algorithms used are supervised and need a large quantity of labeled data to achieve high accuracy. Also, most applications of machine learning in medicine are attempts to mimic or exceed human diagnostic capabilities but little work has been done to show the power of these algorithms to help collect and pre-process a large amount of data. In this study we show how unsupervised learning can extract and sort usable data from large unlabeled datasets with minimal human intervention. Our digital examination tools used in clinical practice store such databases and are largely under-exploited. We applied unsupervised algorithms to corneal topography examinations which remains the gold standard test for diagnosis and follow-up of many corneal diseases and refractive surgery screening. We could extract 7019 usable examinations which were automatically sorted in 3 common diagnoses (Normal, Keratoconus and History of Refractive Surgery) from an unlabeled database with an overall accuracy of 96.5%. Similar methods could be used on any form of digital examination database and greatly speed up the data collection process and yield to the elaboration of stronger supervised models.

Corneal Topography Raw Data Classification Using a Convolutional Neural Network.

PURPOSE: We investigated the efficiency of a convolutional neural network applied to corneal topography raw data to classify examinations of 3 categories: normal, keratoconus (KC), and history of refractive surgery (RS). DESIGN: Retrospective machine-learning experimental study. METHODS: A total of 3,000 Orbscan examinations (1,000 of each class) of different patients of our institution were selected for model training and validation. One hundred examinations of each class were randomly assigned to the test set. For each examination, the raw numerical data from "elevation against the anterior best fit sphere (BFS)," "elevation against the posterior BFS" "axial anterior curvature," and "pachymetry" maps were used. Each map was a square matrix of 2,500 values. The 4 maps were stacked and used as if they were 4 channels of a single image.A convolutional neural network was built and trained on the training set. Classification accuracy and class wise sensitivity and specificity were calculated for the validation set. RESULTS: Overall classification accuracy of the validation set (n = 300) was 99.3% (98.3%-100%). Sensitivity and specificity were, respectively, 100% and 100% for KC, 100% and 99% (94.9%-100%) for normal examinations, and 98% (97.4%-100%) and 100% for RS examinations. CONCLUSION: Using combined corneal topography raw data with a convolutional neural network is an effective way to classify examinations and probably the most thorough way to automatically analyze corneal topography. It should be considered for other routine tasks performed on corneal topography, such as refractive surgery screening.

Retinal and Choroidal Microvasculature in Nonarteritic Anterior Ischemic Optic Neuropathy: An Optical Coherence Tomography Angiography Study.

Purpose: To analyze retinal and choroidal microvasculature in patients with nonarteritic anterior ischemic optic neuropathy (NAION) by using optical coherence tomography angiography (OCT-A). Methods: In this case-control retrospective observational study, patients with atrophic NAION (at least 3 months after onset of symptoms) and normal subjects underwent a complete ophthalmic examination including spectral-domain OCT, visual field (VF), and OCT-A. Whole en face image vessel density (wiVD) was used to assess retinal blood flow of the radial peripapillary capillaries (RPCs), circumpapillary RPC vessel density (cpVD), superficial capillary plexus (SCP), deep capillary plexus (DCP), and choriocapillaris (CC). Statistical correlations between wiVD measurements and visual acuity, VF parameters, retinal nerve fiber layer (RNFL), and combined thickness of retinal ganglion cell and inner plexiform layers were analyzed. Results: Twenty-four patients (26 eyes) with NAION and 24 age-matched normal controls (NCs) (24 eyes) were included. OCT-A showed significant reduction of the RPC wiVD (P < 0.0001) and the cpVD (P < 0.0001) in NAION eyes compared with NC and correlated with RNFL thickness (P = 0.002, P = 0.004), visual acuity (P = 0.042), and mean deviation of the VF (P = 0.001). Macular OCT angiograms showed capillary rarefaction in the SCP (P < 0.0001) and DCP (P < 0.0001) in the NAION group, both correlated with visual acuity (P = 0.02, P = 0.024). However, wiVD of the CC was not significantly different between the two groups in the peripapillary (P = 0.218) and macular (P = 0.786) areas. Conclusions: OCT-A provided detailed visualization of the peripapillary and macular retinal capillary rarefaction, correlated with VF and visual acuity loss. OCT-A could be a useful tool for quantifying and monitoring ischemia in NAION.

Unusual Internal Astigmatism Due to Severe Capsule Contraction Syndrome.

PURPOSE: To describe a case of severe capsule contraction syndrome causing significant astigmatism due to intraocular lens (IOL) folding. METHODS: Case report. RESULTS: Capsule fibrosis and contraction resulted in bending of the hydrophilic IOL along its long axis. Anterior Nd:YAG capsulotomy resolved the situation and restored the patient's visual acuity. CONCLUSIONS: Hydrophilic IOLs are more susceptible to capsule contraction due to the increased flexibility of highly hydrated materials. They should be avoided in patients at risk of capsule contraction to prevent ocular complications. [J Refract Surg. 2018;34(1):65-67.].

Effect of Surgical Intraocular Pressure Lowering on Peripapillary and Macular Vessel Density in Glaucoma Patients: An Optical Coherence Tomography Angiography Study.

PURPOSE: To measure the influence of surgically induced intraocular pressure lowering on peripapillary and macular vessel density in glaucoma patients using optical coherence tomography angiography. METHODS: Twenty-one eyes of 21 patients with open-angle glaucoma scheduled for filtering surgery were enrolled prospectively. Using optical coherence tomography angiography, vessel density was quantified within the peripapillary and macular regions, before and 1 month after filtering surgery. Change in vessel density was calculated for all analyzed areas. RESULTS: One month after surgery, the mean intraocular pressure reduction was 44.2%±4.8% (range, 15.2% to 77.1%). The mean change in vessel density for the whole peripapillary area was 0.065±0.88% (P=0.788). In the macular region, the mean change in vessel density was -0.022%±0.691% (P=0.405) with significant changes only within the inferotemporal area of patients with predominantly superior visual field defects (-1.86%±1.43%, P=0.024). CONCLUSIONS: Optical coherence tomography angiography allowed very limited measurement of intraocular pressure lowering-induced changes on the vessel density of the peripapillary and macular regions in glaucoma patients.

A Severe Case of Pigmentary Glaucoma in a Child With a Family History of Pigment Dispersion Syndrome.

PURPOSE: To report a case of severe pigmentary glaucoma (PG) in a 13-year-old boy of a family affected by pigment dispersion syndrome (PDS). PATIENTS AND METHODS: A 13-year-old child was referred to our hospital for severe bilateral glaucoma. A complete ophthalmologic evaluation including refraction, intraocular pressure, central corneal thickness, slit-lamp biomicroscopy, gonioscopy, fundus examination, and ultrasound biomicroscopy was performed. Family members were also examined and a family pedigree was obtained. RESULTS: Ophthalmologic examination revealed a severe bilateral PG with Krukenberg spindle and a widely open heavily pigmented iridocorneal angle. Ultrasound biomicroscopy showed a deep anterior chamber with pronounced iris concavity in both eyes. Within his family, his 15-year-old sister and 7-year-old brother were both affected by PDS diagnosed on gonioscopy findings. CONCLUSIONS: We report for the first time a severe case of pediatric PG with a family history of PDS. This case demonstrates that accurate screening is necessary in cases of familial PDS and PG, even in the pediatric population.

Optic Disc Vascularization in Glaucoma: Value of Spectral-Domain Optical Coherence Tomography Angiography.

Purpose. To detect changes in optic nerve head (ONH) vascularization in glaucoma patients using spectral-domain OCT angiography (OCT-A). Material and Method. Fifty glaucoma patients and 30 normal subjects were evaluated with OCT-A (AngioVue®, Optovue). The total ONH vessel density and temporal disc vessel density were measured. Clinical data, visual field (VF) parameters, and spectral-domain OCT evaluation (RNFL: retinal nerve fiber layer thickness, GCC: ganglion cell complex thickness, and rim area) were recorded for glaucoma patients. Correlations among total and temporal ONH vessel density and structural and VF parameters were analyzed. Results. In the glaucoma group, total and temporal ONH vessel density were reduced by 24.7% (0.412 versus 0.547; p < 0.0001) and 22.88% (0.364 versus 0.472; p = 0.001), respectively, as compared with the control group. Univariate analysis showed significant correlation between rim area (mm(2)) and temporal ONH vessel density (r = 0.623; p < 0.0001) and total ONH vessel density (r = 0.609; p < 0.0001). Significant correlations were found between temporal and total ONH vessel density and RNFL, GCC, VF mean deviation, and visual field index. Conclusion. In glaucoma patients OCT-A might detect reduced ONH blood vessel density that is associated with structural and functional glaucomatous damage. OCT-A might become a useful tool for the evaluation of ONH microcirculation changes in glaucoma.